Design loop: verify file references, require acceptance tests for methodology claims

[gnovak]

Third PR in the trio addressing bridge-analysis PR #438's failure modes.

• PR Resolve prompt: methodology fidelity, end-to-end verification, write tests #631 hardened resolve (methodology faithfulness, end-to-end verification, write tests for new code, softened anti-exploration language)
• PR Spec generation: verify file refs, ban placeholders, require methodology tests #632 hardened spec generation (verify file refs, ban `...` placeholders, require named methodology tests)
• This PR hardens the design loop — the earliest point where methodology claims and file references first get written down

Why fix this here too

If the design hallucinates "the existing logic lives in module X" when it's actually in `notebooks/Y`, that error propagates: into the spec, into the implementation, eventually into a shipped stub. Catching it at the design stage is cheaper than catching it in resolve. The trio of PRs catches the failure mode at every layer it could be caught.

DEFAULT_SYSTEM_PROMPT additions

• Verify every file/function reference before citing it. Use grep or read_file. If the canonical impl lives in a notebook, say so explicitly — don't pretend it's in a clean module.
• Methodology claims need named acceptance tests. Any "uses BT+EB" / "applies BH-FDR" claim in the design must come with a proposed acceptance test in the analysis (e.g., `test_leaderboard_matches_bt_eb_reference_within_tolerance` with a tolerance value). Without it, the methodology is a label not a contract.

Test plan

• 4 new tests in `TestDesignPromptFidelityRules`
• 676 unit tests total pass
• Empirical: next `/agent-design` or `/agent-delegate` should produce designs that name file paths precisely and include acceptance tests for methodology claims

🤖 Generated with Claude Code